1 INTRODUCTION
Snort is the Network Intrusion Detection System for
rule matching on network data. When detecting
network data, comparing data in the repository is
needed for each match. Therefore, the realization of
Network Intrusion Detection System requires many
comparisons on string operations. If system has
weak computing capabilities, packets’ loss will
occur. In the meantime, the under-reporting network
intrusion will happen. Especially in heavy traffic,
high-load network environment, the package’s false
negative rate is quite higher. Resulting for this
phenomenon mainly comes from as follows:
a) In high throughput network environment, due
to limitations of the hardware and the NIC driver,
improving speed for receiving data is difficult after
reaching a certain peak. Once data stream skyrockets
on the network, the module for receiving data has to
discard many packets. Therefore, incomplete data
and omission of intrusion comes up.
b) The time of detecting one network packet’s
attribution is a little bit long, which is decided by the
pattern matching algorithms. On the one hand, if one
packet is detected by a long time, subsequent
packets will be lately received. Hence, packages’
discard and under-reporting intrusion can happen.
On the other hand, in reality, the intrusions cannot
be disguised with one packet, but many packets
correlate. Discarding any one of them can cause the
false negative.
Therefore, many implementations for Network
Intrusion Detection System utilize multi-core
processor. Teams such as Snort, Haagdorens, Intel
and Giorgos, have developed and tested for Snort on
multi-core platform.
Recently, the multi-core processing platform not
only provides a good environment for the Network
Intrusion Detection System, but also puts forward to
high demands on the parallel Network Intrusion
Detection System.
This paper implements the parallelization for
Network Intrusion Detection System on
high-performance multi-core processing Tile64
platform. First, the efficient capability for network
data reception guarantees the integrity of data input
in Network Intrusion Detection Systems. Second,
according to rules’ requirements for the Network
Intrusion Detection System, the rules are statically
planned. Thus, system distributes the same protocol
Research and Design of Network Intrusion Detection System Based on Multi-Core Tilera64
Liang Zhang & Peiyi Shen & Luobing Dong & Juan Song & Yong Feng & Qiangqiang Liu
XiDian University, Xi’an, Shan Xi, China
Kang Yi
Xi’an Communication Institute, Xi’an, Shan Xi, China
Lukui Zhi & Wentao Zhao
The Public Security Bureau of Shaanxi Province
ABSTRACT: In high-speed network environment, influence factors for the general Network Intrusion
Detection System’s performance mainly include the capability of network packets capturing and the speed of
data processing. According to the issue, this paper designs a parallel Network Intrusion Detection System
based on Multi-Core Tilera64. First, using pipeline decomposition and protocol flow distribution, we deal
with the process of high-speed data capture. Second, based on different configuration requirements for the
intrusion detection, we come up with the distribution strategy for data assignment. At last, a palatalization and
verification system for Snort is realized on Multi-Core Tilera64 platform. From the evaluation, we find that
the system has strong scalability, and its overall performances can be rising with the number of distribution
tiles increasing.
KEYWORDS: Snort; Network Intrusion Detection System; Multi-Core Tilera64; High-Speed Network